Scale Venture Partners - The future of biological foundation models and value creation in AI-driven drug discovery - May 2025
FreeBiological foundation models and value creation in AI drug discovery
FreeFree tier
About Scale Venture Partners - The future of biological foundation models and value creation in AI-driven drug discovery - May 2025
This article from Scale Venture Partners analyzes the potential of biological foundation models (like AlphaFold and ProGen3) to transform AI-driven drug discovery. It discusses the scaling laws observed in these models, compares them to natural language processing, and examines the historical challenges of software in biotech—where value has traditionally accrued to drug assets rather than platforms. The piece offers a balanced view on whether this AI wave represents a genuinely new commercial opportunity or a repeat of past cycles, citing examples such as Schrödinger and the sequencing boom.
Key Features
Analysis of biological foundation models (e.g., AlphaFold, ProGen3)
Discussion of compute-optimal scaling for biological sequences
Historical context of software in drug development (Schrödinger, Celera)
Examination of AI's potential vs. asset-centric biotech economics
Insight into investor perspectives on AI biotech startups
Pros & Cons
Pros
- Provides a balanced, historically-informed analysis of AI in drug discovery
- Cites specific models and companies (AlphaFold, ProGen3, Schrödinger)
- Addresses both technical potential and business/economic realities
- Relevant for investors, strategists, and technology observers
Cons
- Not a practical tool or software—purely an analytical article
- Limited technical depth on model architectures or implementations
- Focuses on investment perspective rather than hands-on scientific guidance
Best For
Understanding the landscape of AI in drug discoveryEvaluating investment opportunities in AI-driven biotechLearning about scaling laws in biological foundation modelsGaining historical perspective on software value capture in life sciences
FAQ
What are biological foundation models?
Biological foundation models are AI models (often transformer-based) trained on biological sequences such as nucleotides (RNA/DNA) or amino acids (proteins) to understand, model, or generate these molecules. Examples include AlphaFold and Profluent's ProGen3.
How does this AI wave differ from earlier tech in drug discovery?
The article notes that unlike past waves (e.g., genome sequencing, earlier AI), current biological foundation models exhibit scaling laws similar to large language models, potentially enabling more generalizable predictions. However, it cautions that biotech's asset-centric economics and long, risky return horizons remain major challenges.
What is the main thesis of the article?
The main thesis questions whether AI represents a fundamentally new commercial opportunity for scientific software in biotech, or whether value will once again accrue to drug assets rather than platform software.